standing jobs · dated deltas常设任务 · 具日期增量

The Discovery Log发现日志

The discovery queues — embedding outliers and formula discovery — are re-run on a standing schedule against the current corpus snapshots. Each run's delta is logged here: which leads entered a queue, which left. A lead that persists across runs and snapshots is more interesting than one that flickers. 两个发现队列 —— 嵌入离群点与套语发现 —— 按常设计划对当前语料快照重跑。每次运行的增量记录于此：哪些线索入列、哪些出列。跨运行、跨快照持续存在的线索，远比一闪而过者更值得留意。

⚠ Every lead in these queues is model-only or derived, and unreviewed until a person looks. Re-run locally: python3 scripts/run_discovery_jobs.py, then rebuild the dataset manifests so published checksums stay true.⚠ 队列中每条线索均为纯模型或派生信号，未经人工审核。本地重跑：python3 scripts/run_discovery_jobs.py，随后重建数据集清单以保持已发布校验和为真。